Projecting COVID-19 Mortality as States Relax Nonpharmacologic Interventions

Key Points

Question  What is the expected trend in COVID-19 mortality if US states were to lift nonpharmacologic interventions (NPIs) at different times over the remainder of 2022?

Findings  In this simulation modeling study, lifting NPIs was likely to result in rebounding epidemics regardless of the delay in lifting. The degree of population-level immunity was associated with the size of the rebounding peak in incident deaths.

Meaning  This simulation study found no path to the end of the COVID-19 pandemic that avoided difficult trade-offs between prolonged NPIs and increased COVID-19 mortality following their removal.

Importance  A key question for policy makers and the public is what to expect from the COVID-19 pandemic going forward as states lift nonpharmacologic interventions (NPIs), such as indoor mask mandates, to prevent COVID-19 transmission.

Objective  To project COVID-19 deaths between March 1, 2022, and December 31, 2022, in each of the 50 US states, District of Columbia, and Puerto Rico assuming different dates of lifting of mask mandates and NPIs.

Design, Setting, and Participants  This simulation modeling study used the COVID-19 Policy Simulator compartmental model to project COVID-19 deaths from March 1, 2022, to December 31, 2022, using simulated populations in the 50 US states, District of Columbia, and Puerto Rico. Projected current epidemiologic trends for each state until December 31, 2022, assuming the current pace of vaccination is maintained into the future and modeling different dates of lifting NPIs.

Exposures  Date of lifting statewide NPI mandates as March 1, April 1, May 1, June 1, or July 1, 2022.

Main Outcomes and Measures  Projected COVID-19 incident deaths from March to December 2022.

Results  With the high transmissibility of current circulating SARS-CoV-2 variants, the simulated lifting of NPIs in March 2022 was associated with resurgences of COVID-19 deaths in nearly every state. In comparison, delaying by even 1 month to lift NPIs in April 2022 was estimated to mitigate the amplitude of the surge. For most states, however, no amount of delay was estimated to be sufficient to prevent a surge in deaths completely. The primary factor associated with recurrent epidemics in the simulation was the assumed high effective reproduction number of unmitigated viral transmission. With a lower level of transmissibility similar to those of the ancestral strains, the model estimated that most states could remove NPIs in March 2022 and likely not see recurrent surges.

Conclusions and Relevance  This simulation study estimated that the SARS-CoV-2 virus would likely continue to take a major toll in the US, even as cases continued to decrease. Because of the high transmissibility of the recent Delta and Omicron variants, premature lifting of NPIs could pose a substantial threat of rebounding surges in morbidity and mortality. At the same time, continued delay in lifting NPIs may not prevent future surges.

The emergency authorization and dissemination of SARS-CoV-2 vaccines starting in late 2020 fundamentally changed the epidemiology of the COVID-19 pandemic. Leading up to the summer of 2021, nearly every state enjoyed falling case rates while simultaneously relaxing nonpharmacologic interventions (NPIs), such as mask mandates and restrictions on social gatherings. This dissociation of social mobility and COVID-19 case rates was a categorical shift that implied a pending end to the pandemic. Unfortunately, around this time, the Delta variant entered circulation in the US and quickly became the dominant SARS-CoV-2 strain. The period of falling case rates was ended by the “fourth wave,” even in states that had achieved relatively high levels of vaccination. Next, even as the Delta variant was still spreading rapidly, the Omicron surge arrived, driving case rates to the highest levels seen in the pandemic and forcing some jurisdictions to reinstate mitigation measures.

The return of NPIs and ongoing need to integrate COVID-19 risk into everyday decision-making have greatly added to pandemic fatigue in the US. Now, as the Omicron wave begins to recede, many states are once again lifting mandatory NPIs, including indoor capacity limits and guidance on social distancing. In particular, local decision makers face the difficult decision of when to lift mask mandates. One of the most important questions currently on the minds of citizens, public health officials, and policy makers is: when can we safely lift restrictions?

We used the COVID-19 Policy Simulator,¹ a compartmental model of SARS-CoV-2 transmission and COVID-19 disease in the US, to project rates of hospitalization and death over the course of the 2022 calendar year assuming different dates of lifting NPIs.

Our model is an extension of the traditional SEIR model, which partitions a population into compartments representing mutually exclusive disease states: susceptible, exposed, infected, recovered, and deceased. In this model, the flow of people between compartments was assumed to obey a system of deterministic ordinary differential equations. The time step was set 1 day to be compatible with data sources reporting daily data. The model was calibrated to historical trends in daily incident deaths up to February 20, 2022. The Table displays the values and data sources for select model parameters. A full model specification is provided in the eAppendix in the Supplement. Where applicable, we followed the Consolidated Health Economic Evaluation Reporting Standards (CHEERS) framework for communicating results. The study was exempt from institutional review board review because it used only publicly available data and was not human participants research.

We stratified the population into 2 age groups: younger than 65 years (lower risk) and 65 years and older (higher risk), assuming that the size of these subpopulations was constant over the simulation period. Age stratification allowed the model to capture differential vaccination trends and COVID-19 mortality between the age groups.

We estimated COVID-19–associated mortality using an infection fatality rate (IFR). The IFR is age specific: 0.5% for the younger than 65 years age group and 3.0% for the 65 years and older age group.

To reflect 2-dose administration guidelines of the COVID-19 messenger RNA vaccines, we stratified the disease states by vaccination status: 0 doses (unvaccinated), 1 dose (partially vaccinated), and 2 doses (fully vaccinated). The third vaccine, the viral vector vaccine, approved for a single-dose regimen, was omitted from the model owing to its accounting for only 3.7% of all administered doses in the US as of June 2021.¹¹ Because there are no data on vaccination status at the time of infection, we assumed doses were allocated proportionally to the susceptible and recovered compartments over the historical time horizon. The vaccine reduces both susceptibility to infection and mortality risk. After the first and second vaccine doses, the probability of contracting the virus was reduced by 46% and 92%, respectively; similarly, the IFR was reduced by 48% and 37%, respectively (see eAppendix in the Supplement for derivation). Note that it is the conditional probability of death that is higher after the second dose than after the first dose, such that the overall reduction in COVID-19 mortality is 72% and 95%, respectively. Vaccine effectiveness was assumed to decrease as the Delta and Omicron variants enter circulation. We defined effective immunity as the sum of the proportion of the populations in the unvaccinated, partially vaccinated, and fully vaccinated susceptible states, weighted by their susceptibility to infection.

For a susceptible individual, the rate of exposure to the virus was dependent on the individual’s risk group, vaccination status, time-varying effective reproduction number, and the size of the infected subpopulation. We estimated an age-stratified matrix of contact patterns among and between the age groups (Table).

An individual who has recovered from natural infection would experience a period of natural immunity before transitioning back into the susceptible state. A fully vaccinated susceptible individual would be protected for the duration of vaccine-conferred immunity before transitioning back into the partially susceptible state. Finally, because individuals with natural immunity who are subsequently vaccinated have been reported to exhibit “unusually potent immune responses,”¹² a fully vaccinated recovered individual was assumed to possess 2 “layers” of immunity, shedding first their natural immunity then their vaccine-conferred immunity. At present, there are no certain estimates of the mean duration of natural and vaccine-conferred immunity. We used the results of a study that examined the immune responses to evolutionarily similar viruses to estimate their time to reinfection under endemic conditions.⁶ Reinfection by endemic SARS-CoV-2 was expected to occur between 3 months and 5 years after peak antibody response, with a median of 16 months.

It was assumed that, once vaccinated, an individual would never shed their immunity completely (within the time frame of the simulation), and a fully vaccinated individual who has shed their vaccine-conferred immunity would be indistinguishable from a partially vaccinated individual. Thus, the model differentiated between the subpopulation that was willing to receive booster shots and the subpopulation that was unwilling to be vaccinated. Fully vaccinated individuals would wane into the partially vaccinated state and would be “boosted” back into the fully vaccinated state.

We projected epidemiologic trends assuming that current rates of infection and vaccination would continue until the date of lifting NPIs, which are the beginning of each calendar month from March to July 2022. We allowed the most recent calibrated value of the effective reproduction number to persist until the lifting date, after which it was increased to the assumed value of 5.0, similar to the basic reproduction number of the Delta variant, representing unmitigated transmission of the virus. We also present projections assuming a lower value of the effective reproduction number of 3.0, similar to the transmissibility of the ancestral strains.⁷

The primary outcome was projections of COVID-19 incident deaths during the remainder of the 2022 calendar year in the 50 US states, the District of Columbia, and Puerto Rico. In addition, we calculated the Spearman rank correlation coefficient as a measure of the correlation between population-level immunity at the time of lifting NPIs and the height of the rebounding surges in COVID-19 deaths. Analyses were performed in February and March 2022 using R, version 4.0.3 (R Foundation for Statistical Computing).

Simulation outcomes indicate that in almost every state, lifting NPIs in 2022 would lead to a substantial rebound in COVID-19 deaths, with peak incident deaths rivaling those seen at the peak of the Omicron surge if lifting occurred in March 2022. Beyond that, however, delaying lifting would not benefit every state equally. In California, Montana, North Carolina, and Oregon, incremental 1-month delays in lifting was estimated to mitigate the amplitude of the rebounding peak in incident deaths. In contrast, the predicted peaks in Florida, Illinois, Michigan, Tennessee, and Washington are similar in size regardless of the timing of lifting, indicating that prolonging restrictions would not meaningfully reduce the disease burden. Moreover, in Massachusetts, New Jersey, New York, and Ohio, delaying lifting was estimated to increase the subsequent peak in incident deaths. Panel A in the Figure presents projections for select states in each group of qualitative outcomes. A complete set of state projections can be found in the eAppendix in the Supplement.

Heterogeneity in the magnitude of the resurgent epidemic across states was driven by the assumption of a single value of the effective reproduction number when no NPIs were in place. It is plausible that smaller states would never reach the level of transmission implied by an effective reproduction number of 5.0 owing to lower population density and activity; hence, the results may be overestimating the severity of their outlook. Panel B in the Figure presents projections with a lower reproduction number of 3.0, in which case the majority of states could lift restrictions with minimal COVID-19 repercussions.

Heterogeneity in response was also associated with differential levels of immunity from both natural infection and vaccination. The combination of waning immunity and falling rates of infection and vaccination means that the net change in population-level immunity would eventually become negative, such that longer delays in lifting could be associated with larger rebounding epidemics. Therefore, current levels of immunity are crucial determinants of the outcomes of returning to higher levels of transmission. We define effective immunity as the sum of the proportion of the populations in the unvaccinated, partially vaccinated, and fully vaccinated susceptible states, weighted by their susceptibility to infection. The Spearman rank correlation coefficient between peak incident deaths (as a percentage of the total population) following the lifting of restrictions on March 1, 2022, and effective immunity on February 20, 2022, was −0.88 (P < 2.2⁻¹⁶). This highly significant and strongly negative correlation suggests that immunity to infection may be associated with a reduction in the severity of the ensuing epidemic following the lifting of NPIs.

We used the COVID-19 Policy Simulator to forecast the number of COVID-19 deaths in each of the 50 US states, the District of Columbia, and Puerto Rico through the 2022 calendar year pending relaxation of NPIs. This analysis could potentially aid state public health officials in evaluating the costs and benefits of lifting NPIs and the timing thereof.

The analysis demonstrates the importance of the timing of lifting NPIs. Premature lifting was estimated to result in recurrent epidemic surges in every almost state. At the same time, a delay of even 1 month was estimated to result in marked reductions to the peak of the mortality curve and the burden on US hospitals. Unfortunately, in most states, no critical moment was identified after which it would be possible to lift NPIs without expecting to see a rebounding surge in deaths. The message that there is no “magic moment” to lift restrictions is important for both sides of the current masking debates in the US. Those opposed to mask mandates should recognize the adverse health outcomes related to relaxing transmission mitigation measures. Any argument to remove such restrictions must address the trade-off and explicitly argue for lifting restrictions within a cost–benefit framework examining the cost of restrictions vs the cost of COVID-19 mortality. At the same time, those who favor maintaining NPIs must recognize that “just a little longer” will not suffice. There is likely no amount of additional waiting time in any state after which removing NPIs will not lead to a rise in morbidity and mortality. The same logic and goals that drive mitigation today will persist, emphasizing the need for mitigation in the future.

A difficult trade-off lies on the horizon. The decision need not be made today, and there is ample evidence that a March 2022 lifting date would have been too soon in many states. However, whenever states do remove NPIs, they will face the same difficult decision regarding the trade-off between increased COVID-19 mortality and the freedoms of returning to a prepandemic norm.

We also estimate that the highly transmissible Delta and Omicron variants will likely continue to take a major toll on the US. The simulations reveal that it is the high transmissibility of these recent variants that sustains the pandemic. With a lower level of transmission similar to that of the ancestral strains, the burden of rebounding morbidity and mortality would be substantially lower. Were this the case, it would likely be possible to remove NPIs at the beginning of the second quarter of 2022.

This study has important limitations. First, we use simulation modeling, which includes all caveats that past performance does not ensure future performance. The COVID-19 Policy Simulator closely replicates historical trends in COVID-19 cases and deaths in all states, but it cannot forecast trends introduced by entirely new dynamics, such as new SARS-CoV-2 variants. Second, the true level of transmission following the lifting of NPIs is uncertain but is obviously a key driver of the outcomes of the analysis. We have presented outcomes with a pessimistic and an optimistic value of the effective reproduction number to allow readers to make their own assessments. Third, the model does not incorporate interstate travel. Predictions may be biased in states that typically experience a high level of travel from other states that have differing levels of COVID-19 cases. Fourth, the model assumes that when NPIs are removed, the virus returns to the level of transmissibility expected in the complete absence of mitigation measures. In reality, individuals may voluntarily continue to wear masks and practice social distancing, which could mitigate the severity of rebounding epidemics.

This study used simulation modeling to project COVID-19 deaths in the each of the 50 US states, the District of Columbia, and Puerto Rico assuming different timing of the lifting of NPIs. We estimated substantial heterogeneity in outcomes between states, that the timing of lifting NPIs is important, that even short delays in lifting could have a big impact, but that there is likely no amount of delay after which it would be completely safe to remove NPIs. Policy makers should consider the findings of this analysis as they monitor their state’s progress during the COVID-19 pandemic, project a suitable time to end restrictions, begin to discuss the conditions that must be met before declaring the pandemic over, and keep the public informed by making public health plans both safe and explicit. Ongoing vaccination efforts will help to contain the COVID-19 pandemic in 2022.

Open Access: This is an open access article distributed under the terms of the CC-BY License. © 2022 Linas BP et al. JAMA Health Forum.

Accepted for Publication: March 8, 2022.

Published: April 1, 2022. doi:10.1001/jamahealthforum.2022.0760

Corresponding Author: Benjamin P. Linas, MD, MPH, Boston Medical Center, 801 Massachusetts Ave, Crosstown Bldg, Room 2007, Boston, MA 02118 (benjamin.linas@bmc.org).

Author Contributions: Ms Xiao had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Dr Linas and Ms Xiao contributed equally to the analysis and are co–first authors. Drs Ayer and Chhatwal contributed equally to the analysis and are co–senior authors.

Concept and design: Linas, Xiao, Mueller, Ayer, Chhatwal.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: Linas, Xiao, Adee, Ayer, Chhatwal.

Critical revision of the manuscript for important intellectual content: Linas, Xiao, Dalgic, Mueller, Aaron, Ayer, Chhatwal.

Statistical analysis: Linas, Xiao, Mueller, Adee, Ayer, Chhatwal.

Obtained funding: Chhatwal.

Administrative, technical, or material support: Dalgic, Adee, Aaron.

Supervision: Linas, Ayer, Chhatwal.

Conflict of Interest Disclosures: Dr Linas reported receiving grants from the National Institute on Drug Abuse, the US Centers for Disease Control and Prevention, and the National Science Foundation during the conduct of the study. Mr Dalgic reported that Value Analytics Labs provided consulting services for Janssen Pharmaceuticals outside the submitted work. Dr Ayer reported being a managing partner at Value Analytics Labs, a health care analytics firm, outside the submitted work. Dr Chhatwal reported receiving grants from the National Science Foundation and Rockefeller Foundation during the conduct of the study; and receiving personal fees from Value Analytics Labs outside the submitted work. No other disclosures were reported.

Funding/Support: This work was supported by National Science Foundation awards 2035360 and 2035361, and the Rockefeller Foundation COVID-19 Modeling Accelerator.

Role of the Funder/Sponsor: The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Additional Information: Statistical code is available from the authors on request to the corresponding author.